Kick. Ass.

I say a few dabs of super-glue and forego the wooden frame all together

PCIe cards can draw ~75W from the PCIe socket.  This is how GPUs got by in the olden days without requiring additional 6/8pin power plugs.

Most cards will still utilize this 75W in addition to the 6/8pin plugs, so 4 cards on a single mobo might be drawing ~300W via the PCIe lanes directly.  That's a lot of current to be pulling through your motherboard.

The PCIe extenders with molex power plugs patched in alleviate this load and allow your PSU to supply this 75W directly, which reduces the strain on your ATX 24pin plug and the motherboard itself.  I've read instances of the 12V leads on the ATX 24pin plug literally melting due to the heat produced from the excessive draw.

If you're going to have more than 3-4 cards in a single motherboard, or several really high-power cards (eg 5970, 6990), it's wise to use these adapters.

I would kill to run my rigs at 46F ambient!

They don't at all like the 95F ambient they're sweating in now.

The colder the better really, the crazy nitrogen overclocker types go sub-zero.  You'll be hard pressed to find too cold.

Overheating could definitely cause it, if you aren't monitoring your temps you should start.  The GPU will automatically throttle to reduce temperature.

Burning anything smell is usually a bad indication

The 188W TDP is max, even the most demanding videogames won't sustain that for very long.  Under "normal" use its probably more than adequate.

The gpu alone probably doesn't do too much more than 100W heat output, so redoing the thermal paste might get you where you need to be.

Specs I'm seeing for the arctic l2 pro show a cooling capacity of 100W:

http://www.arctic.ac/en/p/detail?sArticle=23.%3F#techdata

Your 5870 has a TDP of about 180W I believe, with much of that being converted into heat.

I fear that cooler may just be a bit underpowered for a card running at 95%+ load for extended periods.

If you think you may have overdone the thermal paste, it's worthwhile to try it again, but you might want to look for something a bit beefier.

The only difference between PCIe 1x and PCIe 16x is bandwidth, and mining requires very little.  Some of the FPGA miners I've read about use simple serial links for getwork calls, and that's significantly less available bandwidth than a PCIe 1x slot on the mobo.

I'd look for other differences between the cards - make sure GPUz is showing the proper shader counts, gpu/mem speeds are equal, and that you're using the same miner flags.

Power issues wouldn't usually show themselves as reduced hashrates - likelier you'd see the card not responding at all in that case.  Or in a dual-GPU card, maybe one of the GPUs would be inactive, effectively halving your hashrate.

Another good test would be to swap that card into the mobo and move another out to the extender; if you still get the same 140mhash directly connected, and everything else is equal, something is up with your card.

Yes, yes, and yes, surprisingly so!


I've got 3 rigs in a basement closet, each with two cards. One rig was making it a little warm.  Adding the next two made it unbearably HOT.

I installed a vent to pull cool air in from an adjoining hallway, an 80CFM 6 sones bathroom fan in the ceiling venting directly outside (sharing the dryer exhaust vent), and it's still 95F (35C) in the closet on average.  That fan is audible throughout 2/3 of my house alone, not to mention the noise created by the array of fans cooling (hah!) all my cards.

It quickly becomes very difficult to deal with the heat.

My next step is directed ventilation, running a cool air duct directly to the GPUs, and the exhaust pulling heat off the backs of the cards.  It's going to be a real mess of sheet metal and aluminum tubing!

Plan for at least double the heat you think you might produce, and you might be ok.  Noise? You'll have to learn to deal with it, or spend a lot of time tweaking your fans / ventilation systems to be quieter.

Didn't see it here, but don't Crossfire the cards;  remove the jumper if you've got it attached, and ensure it's not enabled in the software.  The aticonfig commands to check this escape me at the moment.

When Crossfired, it may be trying to treat your cards as a single interface, and denying direct requests to the second card.

I don't have a whole lot of experience in this area, but I do know it caused some similar weirdness when I tried enabling it on  my pair of 5830's.

This!  It never occurred to me to buy 6990s before they're branded and labeled.  Significant savings!

I find your post entertaining! Too bad I just exchanged all my BTC

I think Vladmir is somewhere in that relative vicinity, and needs data centers with special power supply just to get enough power to his rigs.  That'd could easily eat your $70k right there.

[edit] To be fair, he didn't specify GPUs exactly, just machines.  Perhaps he's found the magical fountain of cheap FPGAs? 

Stock clocks are 830 / 1300 (1GB XFX 6950 ZNDC)

Getting ~360mhash at 885/760mhz gpu/mem, newest poclbm (phatk kernel) from git,  -v -f10 -w128

I'm all about connecting the green and blacks with duct electrical tape method.  Any spare bit of wire will do, and you can stack 'em up ad-infinitum

Just my 2 cents, which happens to be what I'm willing to spend on that 

What type of cooler are your 5830's using?  If it's a turbine-type cooler with intake on the rear (the end near your drive cages) of the card, I'd position it blowing into the cards there.  You'll be able to turn your GPU fans way down, it may actually extend their lifespan.

If it's a 'blow everywhere' type like the Saffire 5830's that exhaust out both ends of the card, then positioning so it aids the intake fans will have a similar effect.  I hate this cooler design for mining, as it's dumping heat all over the inside of your case.

+1 on researching pos vs neg pressure - normally it wouldn't make a big difference but you're going to be pushing a huge amount of air.  It may be worth trying to setup a postiive pressure system, theoretically allowing the GPUs to more easily pull air through the fans, across the heatsink, and mostly exhausting heat out the back of your case where it belongs. 

Experimentation is really the key, play with positioning until your GPUs get chilly, and fly with that.

Say you had a really big enclosed case that was ~3 cubic feet in volume.

You'd be replacing the air in the case ~80 times every minute at 240cfm.  I'm thinking it doesn't matter much

http://idle.slashdot.org/story/11/07/07/0326213/Lawyer-Attempts-To-Trademark-Bitcoin


This.  We need more fools just like this.

62dBa is *really* loud

I'd be careful with placement, 240cfm is going to disrupt any other fans in your case that aren't as powerful - ie, if you use it as an exhaust your intake fans will be struggling to keep up.  As an intake, your exhaust fans will be struggling to keep up.

I'm running a couple rigs with 2x 1GB XFX 6950s - shaders aren't unlockable on any of them unfortunately, but at 885/760 gpu/mem clocks I get 345mhash/s each.  Sempron single-core 45W CPU and cheap g-skill ram.

Each rig pulls ~410W at the wall.  1000W is lmao.  I got a nice 750W 80+ Corsair PSU expecting more draw, but even adding a third card I won't hit that limit.